fix: prevent telemetry from leaking sensitive info

Prevent sensitive information from being logged in telemetry output by
assigning SecretStr type to sensitive fields. API keys, password from
KV store are now covered. All providers have been converted.

Signed-off-by: Sébastien Han <seb@redhat.com>

llama_stack/providers/remote/inference/anthropic/anthropic.py

@@ -4,6 +4,7 @@
 # This source code is licensed under the terms described in the LICENSE file in
 # the root directory of this source tree.
 
+
 from llama_stack.providers.utils.inference.litellm_openai_mixin import LiteLLMOpenAIMixin
 from llama_stack.providers.utils.inference.openai_mixin import OpenAIMixin
 
@@ -27,7 +28,7 @@ class AnthropicInferenceAdapter(OpenAIMixin, LiteLLMOpenAIMixin):
         LiteLLMOpenAIMixin.__init__(
             self,
             litellm_provider_name="anthropic",
-            api_key_from_config=config.api_key,
+            api_key_from_config=config.api_key.get_secret_value() if config.api_key else None,
             provider_data_api_key_field="anthropic_api_key",
         )
         self.config = config

llama_stack/providers/remote/inference/anthropic/config.py

@@ -6,13 +6,13 @@
 
 from typing import Any
 
-from pydantic import BaseModel, Field
+from pydantic import BaseModel, Field, SecretStr
 
 from llama_stack.schema_utils import json_schema_type
 
 
 class AnthropicProviderDataValidator(BaseModel):
-    anthropic_api_key: str | None = Field(
+    anthropic_api_key: SecretStr | None = Field(
         default=None,
         description="API key for Anthropic models",
     )
@@ -20,7 +20,7 @@ class AnthropicProviderDataValidator(BaseModel):
 
 @json_schema_type
 class AnthropicConfig(BaseModel):
-    api_key: str | None = Field(
+    api_key: SecretStr | None = Field(
         default=None,
         description="API key for Anthropic models",
     )

llama_stack/providers/remote/inference/gemini/config.py

@@ -6,13 +6,13 @@
 
 from typing import Any
 
-from pydantic import BaseModel, Field
+from pydantic import BaseModel, Field, SecretStr
 
 from llama_stack.schema_utils import json_schema_type
 
 
 class GeminiProviderDataValidator(BaseModel):
-    gemini_api_key: str | None = Field(
+    gemini_api_key: SecretStr | None = Field(
         default=None,
         description="API key for Gemini models",
     )
@@ -20,7 +20,7 @@ class GeminiProviderDataValidator(BaseModel):
 
 @json_schema_type
 class GeminiConfig(BaseModel):
-    api_key: str | None = Field(
+    api_key: SecretStr | None = Field(
         default=None,
         description="API key for Gemini models",
     )

llama_stack/providers/remote/inference/gemini/gemini.py

@@ -4,6 +4,7 @@
 # This source code is licensed under the terms described in the LICENSE file in
 # the root directory of this source tree.
 
+
 from llama_stack.providers.utils.inference.litellm_openai_mixin import LiteLLMOpenAIMixin
 from llama_stack.providers.utils.inference.openai_mixin import OpenAIMixin
 
@@ -19,7 +20,7 @@ class GeminiInferenceAdapter(OpenAIMixin, LiteLLMOpenAIMixin):
         LiteLLMOpenAIMixin.__init__(
             self,
             litellm_provider_name="gemini",
-            api_key_from_config=config.api_key,
+            api_key_from_config=config.api_key.get_secret_value() if config.api_key else None,
             provider_data_api_key_field="gemini_api_key",
         )
         self.config = config

llama_stack/providers/remote/inference/groq/config.py

@@ -6,13 +6,13 @@
 
 from typing import Any
 
-from pydantic import BaseModel, Field
+from pydantic import BaseModel, Field, SecretStr
 
 from llama_stack.schema_utils import json_schema_type
 
 
 class GroqProviderDataValidator(BaseModel):
-    groq_api_key: str | None = Field(
+    groq_api_key: SecretStr | None = Field(
         default=None,
         description="API key for Groq models",
     )
@@ -20,7 +20,7 @@ class GroqProviderDataValidator(BaseModel):
 
 @json_schema_type
 class GroqConfig(BaseModel):
-    api_key: str | None = Field(
+    api_key: SecretStr | None = Field(
         # The Groq client library loads the GROQ_API_KEY environment variable by default
         default=None,
         description="The Groq API key",

llama_stack/providers/remote/inference/llama_openai_compat/config.py

@@ -6,13 +6,13 @@
 
 from typing import Any
 
-from pydantic import BaseModel, Field
+from pydantic import BaseModel, Field, SecretStr
 
 from llama_stack.schema_utils import json_schema_type
 
 
 class LlamaProviderDataValidator(BaseModel):
-    llama_api_key: str | None = Field(
+    llama_api_key: SecretStr | None = Field(
         default=None,
         description="API key for api.llama models",
     )
@@ -20,7 +20,7 @@ class LlamaProviderDataValidator(BaseModel):
 
 @json_schema_type
 class LlamaCompatConfig(BaseModel):
-    api_key: str | None = Field(
+    api_key: SecretStr | None = Field(
         default=None,
         description="The Llama API key",
     )

llama_stack/providers/remote/inference/openai/config.py

@@ -6,13 +6,13 @@
 
 from typing import Any
 
-from pydantic import BaseModel, Field
+from pydantic import BaseModel, Field, SecretStr
 
 from llama_stack.schema_utils import json_schema_type
 
 
 class OpenAIProviderDataValidator(BaseModel):
-    openai_api_key: str | None = Field(
+    openai_api_key: SecretStr | None = Field(
         default=None,
         description="API key for OpenAI models",
     )
@@ -20,7 +20,7 @@ class OpenAIProviderDataValidator(BaseModel):
 
 @json_schema_type
 class OpenAIConfig(BaseModel):
-    api_key: str | None = Field(
+    api_key: SecretStr | None = Field(
         default=None,
         description="API key for OpenAI models",
     )

llama_stack/providers/remote/inference/runpod/config.py

@@ -6,7 +6,7 @@
 
 from typing import Any
 
-from pydantic import BaseModel, Field
+from pydantic import BaseModel, Field, SecretStr
 
 from llama_stack.schema_utils import json_schema_type
 
@@ -17,7 +17,7 @@ class RunpodImplConfig(BaseModel):
         default=None,
         description="The URL for the Runpod model serving endpoint",
     )
-    api_token: str | None = Field(
+    api_token: SecretStr | None = Field(
         default=None,
         description="The API token",
     )

llama_stack/providers/remote/inference/runpod/runpod.py

@@ -103,7 +103,10 @@ class RunpodInferenceAdapter(
             tool_config=tool_config,
         )
 
-        client = OpenAI(base_url=self.config.url, api_key=self.config.api_token)
+        client = OpenAI(
+            base_url=self.config.url,
+            api_key=self.config.api_token.get_secret_value() if self.config.api_token else None,
+        )
         if stream:
             return self._stream_chat_completion(request, client)
         else:

llama_stack/providers/remote/inference/vertexai/vertexai.py

@@ -8,6 +8,7 @@ from typing import Any
 
 import google.auth.transport.requests
 from google.auth import default
+from pydantic import SecretStr
 
 from llama_stack.apis.inference import ChatCompletionRequest
 from llama_stack.providers.utils.inference.litellm_openai_mixin import (
@@ -43,7 +44,7 @@ class VertexAIInferenceAdapter(OpenAIMixin, LiteLLMOpenAIMixin):
         except Exception:
             # If we can't get credentials, return empty string to let LiteLLM handle it
             # This allows the LiteLLM mixin to work with ADC directly
-            return ""
+            return SecretStr("")
 
     def get_base_url(self) -> str:
         """

llama_stack/providers/remote/inference/vllm/config.py

@@ -6,7 +6,7 @@
 
 from pathlib import Path
 
-from pydantic import BaseModel, Field, field_validator
+from pydantic import BaseModel, Field, SecretStr, field_validator
 
 from llama_stack.schema_utils import json_schema_type
 
@@ -21,8 +21,8 @@ class VLLMInferenceAdapterConfig(BaseModel):
         default=4096,
         description="Maximum number of tokens to generate.",
     )
-    api_token: str | None = Field(
-        default="fake",
+    api_token: SecretStr | None = Field(
+        default=SecretStr("fake"),
         description="The API token",
     )
     tls_verify: bool | str = Field(

llama_stack/providers/remote/inference/vllm/vllm.py

@@ -294,7 +294,7 @@ class VLLMInferenceAdapter(OpenAIMixin, LiteLLMOpenAIMixin, Inference, ModelsPro
             self,
             model_entries=build_hf_repo_model_entries(),
             litellm_provider_name="vllm",
-            api_key_from_config=config.api_token,
+            api_key_from_config=config.api_token.get_secret_value(),
             provider_data_api_key_field="vllm_api_token",
             openai_compat_api_base=config.url,
         )